diff --git a/python/llm/example/gpu/hf-transformers-models/aquila/README.md b/python/llm/example/gpu/hf-transformers-models/aquila/README.md
new file mode 100644
index 00000000..56a8fc6e
--- /dev/null
+++ b/python/llm/example/gpu/hf-transformers-models/aquila/README.md
@@ -0,0 +1,57 @@
+# Aquila
+
+In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Aquila models. For illustration purposes, we utilize the [BAAI/AquilaChat-7B](https://huggingface.co/BAAI/AquilaChat-7B) as a reference Aquila model.
+
+> **Note**: If you want to download the Hugging Face *Transformers* model, please refer to [here](https://huggingface.co/docs/hub/models-downloading#using-git).
+>
+> BigDL-LLM optimizes the *Transformers* model in INT4 precision at runtime, and thus no explicit conversion is needed.
+
+## Requirements
+To run these examples with BigDL-LLM, we have some recommended requirements for your machine, please refer to [here](../README.md#recommended-requirements) for more information.
+
+## Example: Predict Tokens using `generate()` API
+In the example [generate.py](./generate.py), we show a basic use case for a Aquila model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations.
+
+### 1. Install
+We suggest using conda to manage environment:
+```bash
+conda create -n llm python=3.9
+conda activate llm
+# below command will install intel_extension_for_pytorch==2.0.110+xpu as default
+# you can install specific ipex/torch version for your need
+pip install --pre --upgrade bigdl-llm[xpu] -f https://developer.intel.com/ipex-whl-stable-xpu
+```
+### 2. Configures OneAPI environment variables
+```bash
+source /opt/intel/oneapi/setvars.sh
+```
+
+### 3. Run
+
+For optimal performance on Arc, it is recommended to set several environment variables.
+
+```bash
+export USE_XETLA=OFF
+export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
+```
+
+```
+python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT
+```
+
+Arguments Info
+In the example, several arguments can be passed to satisfy your requirements:
+
+- `--repo-id-or-model-path`: str, argument defining the huggingface repo id for the Aquila model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'BAAI/AquilaChat-7B'`.
+- `--prompt`: str, argument defining the prompt to be inferred (with integrated prompt format for chat). It is default to be `'AI是什么？'`.
+- `--n-predict`: int, argument defining the max number of tokens to predict. It is default to be `32`.
+
+#### Sample Output
+#### [BAAI/AquilaChat-7B](https://huggingface.co/BAAI/AquilaChat-7B)
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+Human: AI是什么？###Assistant:
+-------------------- Output --------------------
+Human: AI是什么？###Assistant: AI是人工智能的缩写。人工智能是一种技术，旨在使计算机能够像人类一样思考、学习和执行任务。AI包括许多不同的技术和方法，例如机器
+```
diff --git a/python/llm/example/gpu/hf-transformers-models/aquila/generate.py b/python/llm/example/gpu/hf-transformers-models/aquila/generate.py
new file mode 100644
index 00000000..d070a75c
--- /dev/null
+++ b/python/llm/example/gpu/hf-transformers-models/aquila/generate.py
@@ -0,0 +1,73 @@
+#
+# Copyright 2016 The BigDL Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+import torch
+import intel_extension_for_pytorch as ipex
+import time
+import argparse
+
+from bigdl.llm.transformers import AutoModelForCausalLM
+from transformers import AutoTokenizer
+
+# you could tune the prompt based on your own model,
+# here the prompt tuning refers to https://huggingface.co/BAAI/AquilaChat-7B/blob/13577616fd4ff0d21c5735a88d350a68dae120e5/cyg_conversation.py
+AQUILA_PROMPT_FORMAT = "Human: {prompt}###Assistant:"
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(
+        description='Predict Tokens using `generate()` API for Aquila model')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="BAAI/AquilaChat-7B",
+                        help='The huggingface repo id for the Aquila model to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--prompt', type=str, default="AI是什么？",
+                        help='Prompt to infer')
+    parser.add_argument('--n-predict', type=int, default=32,
+                        help='Max tokens to predict')
+
+    args = parser.parse_args()
+    model_path = args.repo_id_or_model_path
+
+    # Load model in 4 bit,
+    # which convert the relevant layers in the model into INT4 format
+    model = AutoModelForCausalLM.from_pretrained(model_path,
+                                                 load_in_4bit=True,
+                                                 trust_remote_code=True)
+    model = model.to('xpu')
+
+    # Load tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(model_path,
+                                              trust_remote_code=True)
+
+    # Generate predicted tokens
+    with torch.inference_mode():
+        prompt = AQUILA_PROMPT_FORMAT.format(prompt=args.prompt)
+        input_ids = tokenizer.encode(prompt, return_tensors="pt").to('xpu')
+        st = time.time()
+        # if your selected model is capable of utilizing previous key/value attentions
+        # to enhance decoding speed, but has `"use_cache": false` in its model config,
+        # it is important to set `use_cache=True` explicitly in the `generate` function
+        # to obtain optimal performance with BigDL-LLM INT4 optimizations
+        output = model.generate(input_ids,
+                                max_new_tokens=args.n_predict)
+        torch.xpu.synchronize()
+        end = time.time()
+        output = output.cpu()
+        output_str = tokenizer.decode(output[0], skip_special_tokens=True)
+        print(f'Inference time: {end - st} s')
+        print('-' * 20, 'Prompt', '-' * 20)
+        print(prompt)
+        print('-' * 20, 'Output', '-' * 20)
+        print(output_str)
diff --git a/python/llm/example/gpu/hf-transformers-models/dolly-v1/README.md b/python/llm/example/gpu/hf-transformers-models/dolly-v1/README.md
new file mode 100644
index 00000000..d1a1be39
--- /dev/null
+++ b/python/llm/example/gpu/hf-transformers-models/dolly-v1/README.md
@@ -0,0 +1,67 @@
+# Dolly v1
+In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Dolly v1 models. For illustration purposes, we utilize the [databricks/dolly-v1-6b](https://huggingface.co/databricks/dolly-v1-6b) as a reference Dolly v1 model.
+
+## 0. Requirements
+To run these examples with BigDL-LLM, we have some recommended requirements for your machine, please refer to [here](../README.md#recommended-requirements) for more information.
+
+## Example: Predict Tokens using `generate()` API
+In the example [generate.py](./generate.py), we show a basic use case for a Dolly v1 model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations.
+
+
+### 1. Install
+We suggest using conda to manage environment:
+```bash
+conda create -n llm python=3.9
+conda activate llm
+# below command will install intel_extension_for_pytorch==2.0.110+xpu as default
+# you can install specific ipex/torch version for your need
+pip install --pre --upgrade bigdl-llm[xpu] -f https://developer.intel.com/ipex-whl-stable-xpu
+```
+### 2. Configures OneAPI environment variables
+```bash
+source /opt/intel/oneapi/setvars.sh
+```
+
+### 3. Run
+
+For optimal performance on Arc, it is recommended to set several environment variables.
+
+```bash
+export USE_XETLA=OFF
+export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
+```
+
+```
+python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT
+```
+
+Arguments info:
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Dolly v1 model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'databricks/dolly-v1-6b'`.
+- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'What is AI?'`.
+- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`.
+
+> **Note**: When loading the model in 4-bit, BigDL-LLM converts linear layers in the model into INT4 format. In theory, a *X*B model saved in 16-bit will requires approximately 2*X* GB of memory for loading, and ~0.5*X* GB memory for further inference.
+>
+> Please select the appropriate size of the Dolly v1 model based on the capabilities of your machine.
+
+#### Sample Output
+#### [databricks/dolly-v1-6b](https://huggingface.co/databricks/dolly-v1-6b)
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+Below is an instruction that describes a task. Write a response that appropriately completes the request.
+
+### Instruction:
+What is AI?
+
+### Response:
+
+-------------------- Output --------------------
+Below is an instruction that describes a task. Write a response that appropriately completes the request.
+
+### Instruction:
+What is AI?
+
+### Response:
+AI is an umbrella term for a variety of technologies that enable computers to think and act like humans. AI can be used to automate tasks, analyze data, and
+```
diff --git a/python/llm/example/gpu/hf-transformers-models/dolly-v1/generate.py b/python/llm/example/gpu/hf-transformers-models/dolly-v1/generate.py
new file mode 100644
index 00000000..75c058b3
--- /dev/null
+++ b/python/llm/example/gpu/hf-transformers-models/dolly-v1/generate.py
@@ -0,0 +1,85 @@
+#
+# Copyright 2016 The BigDL Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+import torch
+import intel_extension_for_pytorch as ipex
+import time
+import argparse
+import numpy as np
+
+from bigdl.llm.transformers import AutoModelForCausalLM
+from transformers import AutoTokenizer
+
+# you could tune the prompt based on your own model,
+# here the prompt tuning refers to https://huggingface.co/databricks/dolly-v1-6b#generate-text
+DOLLY_V1_PROMPT_FORMAT = """Below is an instruction that describes a task. Write a response that appropriately completes the request.
+
+### Instruction:
+{prompt}
+
+### Response:
+"""
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Dolly v1 model')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="databricks/dolly-v1-6b",
+                        help='The huggingface repo id for the Dolly v1 model to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--prompt', type=str, default="What is AI?",
+                        help='Prompt to infer')
+    parser.add_argument('--n-predict', type=int, default=32,
+                        help='Max tokens to predict')
+
+    args = parser.parse_args()
+    model_path = args.repo_id_or_model_path
+
+    # Load model in 4 bit,
+    # which convert the relevant layers in the model into INT4 format
+    model = AutoModelForCausalLM.from_pretrained(model_path,
+                                                 load_in_4bit=True)
+    model = model.to('xpu')
+
+    # Load tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(model_path)
+    
+    # Generate predicted tokens
+    with torch.inference_mode():
+        prompt = DOLLY_V1_PROMPT_FORMAT.format(prompt=args.prompt)
+        input_ids = tokenizer.encode(prompt, return_tensors="pt").to('xpu')
+        end_key_token_id=tokenizer.encode("### End")[0]
+        st = time.time()
+        # enabling `use_cache=True` allows the model to utilize the previous
+        # key/values attentions to speed up decoding;
+        # to obtain optimal performance with BigDL-LLM INT4 optimizations,
+        # it is important to set use_cache=True for Dolly v1 models
+        output = model.generate(input_ids,
+                                use_cache=True,
+                                max_new_tokens=args.n_predict,
+                                pad_token_id=tokenizer.pad_token_id,
+                                eos_token_id=end_key_token_id)
+        torch.xpu.synchronize()
+        end = time.time()
+        output = output.cpu()
+        end_token_position = None
+        end_token_positions = np.where(output[0] == end_key_token_id)[0]
+        if len(end_token_positions) > 0:
+            end_token_position = end_token_positions[0]
+        output_str = tokenizer.decode(output[0][:end_token_position], skip_special_tokens=False)
+        print(f'Inference time: {end-st} s')
+        print('-'*20, 'Prompt', '-'*20)
+        print(prompt)
+        print('-'*20, 'Output', '-'*20)
+        print(output_str)
diff --git a/python/llm/example/gpu/hf-transformers-models/dolly-v2/README.md b/python/llm/example/gpu/hf-transformers-models/dolly-v2/README.md
new file mode 100644
index 00000000..7ef54e16
--- /dev/null
+++ b/python/llm/example/gpu/hf-transformers-models/dolly-v2/README.md
@@ -0,0 +1,71 @@
+# Dolly v2
+In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Dolly v2 models. For illustration purposes, we utilize the [databricks/dolly-v2-12b](https://huggingface.co/databricks/dolly-v2-12b) as a reference Dolly v2 model.
+
+## 0. Requirements
+To run these examples with BigDL-LLM, we have some recommended requirements for your machine, please refer to [here](../README.md#recommended-requirements) for more information.
+
+## Example: Predict Tokens using `generate()` API
+In the example [generate.py](./generate.py), we show a basic use case for a Dolly v2 model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations.
+### 1. Install
+We suggest using conda to manage environment:
+```bash
+conda create -n llm python=3.9
+conda activate llm
+
+pip install bigdl-llm[all] # install bigdl-llm with 'all' option
+```
+
+### 2. Run
+```
+python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT
+```
+
+Arguments info:
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Dolly v2 model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'databricks/dolly-v2-12b'`.
+- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'What is AI?'`.
+- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`.
+
+> **Note**: When loading the model in 4-bit, BigDL-LLM converts linear layers in the model into INT4 format. In theory, a *X*B model saved in 16-bit will requires approximately 2*X* GB of memory for loading, and ~0.5*X* GB memory for further inference.
+>
+> Please select the appropriate size of the Dolly v2 model based on the capabilities of your machine.
+
+#### 2.1 Client
+On client Windows machine, it is recommended to run directly with full utilization of all cores:
+```powershell
+python ./generate.py 
+```
+
+#### 2.2 Server
+For optimal performance on server, it is recommended to set several environment variables (refer to [here](../README.md#best-known-configuration-on-linux) for more information), and run the example with all the physical cores of a single socket.
+
+E.g. on Linux,
+```bash
+# set BigDL-Nano env variables
+source bigdl-nano-init
+
+# e.g. for a server with 48 cores per socket
+export OMP_NUM_THREADS=48
+numactl -C 0-47 -m 0 python ./generate.py
+```
+
+#### 2.3 Sample Output
+#### [databricks/dolly-v2-12b](https://huggingface.co/databricks/dolly-v2-12b)
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+Below is an instruction that describes a task. Write a response that appropriately completes the request.
+
+### Instruction:
+What is AI?
+
+### Response:
+
+-------------------- Output --------------------
+Below is an instruction that describes a task. Write a response that appropriately completes the request.
+
+### Instruction:
+What is AI?
+
+### Response:
+Artificial Intelligence (AI) is the area of computer science concerned with building machines that can perform tasks normally associated with human intelligence, such as reasoning, learning,
+```
diff --git a/python/llm/example/gpu/hf-transformers-models/dolly-v2/generate.py b/python/llm/example/gpu/hf-transformers-models/dolly-v2/generate.py
new file mode 100644
index 00000000..d785dec2
--- /dev/null
+++ b/python/llm/example/gpu/hf-transformers-models/dolly-v2/generate.py
@@ -0,0 +1,86 @@
+#
+# Copyright 2016 The BigDL Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+import torch
+import intel_extension_for_pytorch as ipex
+import time
+import argparse
+import numpy as np
+
+from bigdl.llm.transformers import AutoModelForCausalLM
+from transformers import AutoTokenizer
+
+# you could tune the prompt based on your own model,
+# here the prompt tuning refers to https://huggingface.co/databricks/dolly-v2-12b/blob/main/instruct_pipeline.py#L15
+DOLLY_V2_PROMPT_FORMAT = """Below is an instruction that describes a task. Write a response that appropriately completes the request.
+
+### Instruction:
+{prompt}
+
+### Response:
+"""
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Dolly v2 model')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="databricks/dolly-v2-12b",
+                        help='The huggingface repo id for the Dolly v2 model to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--prompt', type=str, default="What is AI?",
+                        help='Prompt to infer')
+    parser.add_argument('--n-predict', type=int, default=32,
+                        help='Max tokens to predict')
+
+    args = parser.parse_args()
+    model_path = args.repo_id_or_model_path
+
+    # Load model in 4 bit,
+    # which convert the relevant layers in the model into INT4 format
+    model = AutoModelForCausalLM.from_pretrained(model_path,
+                                                 load_in_4bit=True,
+                                                 trust_remote_code=True)
+    model = model.to('xpu')
+
+    # Load tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(model_path,
+                                              trust_remote_code=True)
+    
+    # Generate predicted tokens
+    with torch.inference_mode():
+        prompt = DOLLY_V2_PROMPT_FORMAT.format(prompt=args.prompt)
+        input_ids = tokenizer.encode(prompt, return_tensors="pt").to('xpu')
+        end_key_token_id=tokenizer.encode("### End")[0]
+        st = time.time()
+        # if your selected model is capable of utilizing previous key/value attentions
+        # to enhance decoding speed, but has `"use_cache": false` in its model config,
+        # it is important to set `use_cache=True` explicitly in the `generate` function
+        # to obtain optimal performance with BigDL-LLM INT4 optimizations
+        output = model.generate(input_ids,
+                                max_new_tokens=args.n_predict,
+                                pad_token_id=tokenizer.pad_token_id,
+                                eos_token_id=end_key_token_id)
+        torch.xpu.synchronize()
+        end = time.time()
+        output = output.cpu()
+        end_token_position = None
+        end_token_positions = np.where(output[0] == end_key_token_id)[0]
+        if len(end_token_positions) > 0:
+            end_token_position = end_token_positions[0]
+        output_str = tokenizer.decode(output[0][:end_token_position], skip_special_tokens=False)
+        print(f'Inference time: {end-st} s')
+        print('-'*20, 'Prompt', '-'*20)
+        print(prompt)
+        print('-'*20, 'Output', '-'*20)
+        print(output_str)
diff --git a/python/llm/example/gpu/hf-transformers-models/vicuna/README.md b/python/llm/example/gpu/hf-transformers-models/vicuna/README.md
new file mode 100644
index 00000000..0ea9149b
--- /dev/null
+++ b/python/llm/example/gpu/hf-transformers-models/vicuna/README.md
@@ -0,0 +1,76 @@
+# Vicuna
+In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Vicuna models. For illustration purposes, we utilize the [lmsys/vicuna-13b-v1.3](https://huggingface.co/lmsys/vicuna-13b-v1.3) and [eachadea/vicuna-7b-1.1](https://huggingface.co/eachadea/vicuna-7b-1.1) as reference Vicuna models.
+
+## 0. Requirements
+To run these examples with BigDL-LLM, we have some recommended requirements for your machine, please refer to [here](../README.md#recommended-requirements) for more information.
+
+## Example: Predict Tokens using `generate()` API
+In the example [generate.py](./generate.py), we show a basic use case for a Vicuna model to predict the next N tokens using `generate()` API, with BigDL-LLM INT4 optimizations.
+
+
+### 1. Install
+We suggest using conda to manage environment:
+```bash
+conda create -n llm python=3.9
+conda activate llm
+# below command will install intel_extension_for_pytorch==2.0.110+xpu as default
+# you can install specific ipex/torch version for your need
+pip install --pre --upgrade bigdl-llm[xpu] -f https://developer.intel.com/ipex-whl-stable-xpu
+```
+### 2. Configures OneAPI environment variables
+```bash
+source /opt/intel/oneapi/setvars.sh
+```
+
+### 3. Run
+
+For optimal performance on Arc, it is recommended to set several environment variables.
+
+```bash
+export USE_XETLA=OFF
+export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
+```
+
+```
+python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT
+```
+
+Arguments info:
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Vicuna model (e.g. `lmsys/vicuna-13b-v1.3` and `eachadea/vicuna-7b-1.1`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'lmsys/vicuna-13b-v1.3'`.
+- `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'What is AI?'`.
+- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`.
+
+> **Note**: When loading the model in 4-bit, BigDL-LLM converts linear layers in the model into INT4 format. In theory, a *X*B model saved in 16-bit will requires approximately 2*X* GB of memory for loading, and ~0.5*X* GB memory for further inference.
+>
+> Please select the appropriate size of the Vicuna model based on the capabilities of your machine.
+
+#### Sample Output
+#### [lmsys/vicuna-13b-v1.3](https://huggingface.co/lmsys/vicuna-13b-v1.3)
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+### Human:
+What is AI? 
+ ### Assistant:
+
+-------------------- Output --------------------
+### Human:
+What is AI? 
+ ### Assistant:
+AI, or Artificial Intelligence, refers to the development of computer systems that can perform tasks that typically require human intelligence, such as visual perception,
+```
+
+#### [eachadea/vicuna-7b-1.1](https://huggingface.co/eachadea/vicuna-7b-1.1)
+```log
+Inference time: xxxx s
+-------------------- Prompt --------------------
+### Human:
+What is AI? 
+ ### Assistant:
+
+-------------------- Output --------------------
+### Human:
+What is AI? 
+ ### Assistant:
+AI, or artificial intelligence, refers to the ability of a machine or computer program to mimic human intelligence and perform tasks that would normally require human intelligence to
+```
diff --git a/python/llm/example/gpu/hf-transformers-models/vicuna/generate.py b/python/llm/example/gpu/hf-transformers-models/vicuna/generate.py
new file mode 100644
index 00000000..f66d3dce
--- /dev/null
+++ b/python/llm/example/gpu/hf-transformers-models/vicuna/generate.py
@@ -0,0 +1,71 @@
+#
+# Copyright 2016 The BigDL Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+import torch
+import intel_extension_for_pytorch as ipex
+import time
+import argparse
+
+from bigdl.llm.transformers import AutoModelForCausalLM
+from transformers import LlamaTokenizer
+
+# you could tune the prompt based on your own model,
+# here the prompt tuning refers to https://github.com/lm-sys/FastChat/blob/main/docs/vicuna_weights_version.md#example-prompt-weights-v0
+Vicuna_PROMPT_FORMAT = "### Human:\n{prompt} \n ### Assistant:\n"
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Vicuna model')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="lmsys/vicuna-13b-v1.3",
+                        help='The huggingface repo id for the Vicuna (e.g. `lmsys/vicuna-13b-v1.3` and `eachadea/vicuna-7b-1.1`) to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--prompt', type=str, default="What is AI?",
+                        help='Prompt to infer')
+    parser.add_argument('--n-predict', type=int, default=32,
+                        help='Max tokens to predict')
+
+    args = parser.parse_args()
+    model_path = args.repo_id_or_model_path
+
+    # Load model in 4 bit,
+    # which convert the relevant layers in the model into INT4 format
+    model = AutoModelForCausalLM.from_pretrained(model_path,
+                                                 load_in_4bit=True)
+    model = model.to('xpu')
+
+    # Load tokenizer
+    tokenizer = LlamaTokenizer.from_pretrained(model_path)
+    
+    # Generate predicted tokens
+    with torch.inference_mode():
+        prompt = Vicuna_PROMPT_FORMAT.format(prompt=args.prompt)
+        input_ids = tokenizer.encode(prompt, return_tensors="pt").to('xpu')
+        st = time.time()
+        # enabling `use_cache=True` allows the model to utilize the previous
+        # key/values attentions to speed up decoding;
+        # to obtain optimal performance with BigDL-LLM INT4 optimizations,
+        # it is important to set use_cache=True for vicuna-v1.3 models
+        output = model.generate(input_ids,
+                                use_cache=True,
+                                max_new_tokens=args.n_predict)
+        torch.xpu.synchronize()
+        end = time.time()
+        output = output.cpu()
+        output_str = tokenizer.decode(output[0], skip_special_tokens=True)
+        print(f'Inference time: {end-st} s')
+        print('-'*20, 'Prompt', '-'*20)
+        print(prompt)
+        print('-'*20, 'Output', '-'*20)
+        print(output_str)